Now That Large Language Models Are Here, Is Google Screwed?
Technology
Hardly anyone except OpenAI insiders knew how powerful and game-changing ChatGPT would be when it was released on November 30 last year. Even regular AI commentators were shocked by the power of large language models to answer seemingly random questions on just about any topic. The machine seemed to be able to do just about anything, including writing code, answering college essay questions, and churning out blog content that would have cost a fortune just days prior.
It took a couple of weeks for everyone to digest this new Gutenberg moment. It soon became clear that such a tool would change everything, possibly plunging millions into unemployment and putting some of the largest firms in the world out of business.
When ChatGPT was released to the public last year, Google put out a rare “code red” alert to all its employees. The message was simple: the biggest search engine in the world was under threat by a new technology hardly anyone saw coming. OpenAI’s system had the potential to invalidate the need for search altogether, providing users answers to questions that Google used to monetize.
For users, it’s a massive win. ChatGTP is currently ad-free, and the technology provides them with detailed answers to their questions immediately. While there are still some bugs and issues with trustworthiness, there’s no doubting the underlying technology. It has the power to change anything, removing the need for traditional search.
If you haven’t used these new large language model tools yet, you should. The experience is entirely different from a search engine. Instead of trawling for ages through website links to find the information you want, these tools can potentially present it to you on a plate, negating the need to spend time “surfing the web.” You just ask this God-like machine a question, and it spits out an answer.
Right now, the quality of the syntax and written communication is a little simplistic and robotic. It feels a bit like boilerplate text. But it still has immense value. The passion might be lacking, but the information is there.
Again, critics will point to the system’s limitations. But OpenAI is already planning various ChatGTP updates to make the system even more powerful. It’s not clear that anything will stop it from improving continually, short of draconian government regulation.
Search engines are now scrambling to do something about the situation, particularly Google, the market leader, a company worth more than $1 trillion. The firm is desperately trying to counter the threat of AI-driven text and web-based fact-finding, using whatever means it can.
What About Google Bard?
From various corporate releases since last year, it’s clear that Google knew about the threat of large language models. In fact, the California-based tech firm had been developing its own, leveraging innovation from its in-house boffins and Deep Mind. However, the company wasn’t convinced that its LaMDA-powered large language models were up to the task.
What is Google Bard? Essentially, it’s Google’s in-house version of ChatGTP. It uses similar technology to the other tools on the market to provide users with results.
Google has had a working model on the test bench for a couple of years now. However, the company refused to release it to the general public, even after ChatGTP was released.
The reason was simple: risk management. ChatGTP’s answers don’t matter that much because it only serves a few million users. Google, on the other hand, serves billions and is a steward of the world’s information. It can’t afford to trial untested software live to the general public because of the brand damage it could cause. Hence, hardly anyone knew about Bard until Google’s botched demo.
Google Isn’t Stupid
With that said, Google executives aren’t idiots. They know the power of ChatGTP and the risks posed by rival Microsoft using the technology in its search engine results. If Bing can leverage large language models at scale, it could potentially dethrone Google’s current dominant position and replace it with something new.
Microsoft invested billions of dollars in Bing over the last decade in an attempt to gain traction in the search market, but nothing seemed to work. Despite being the default search tool on Windows, nobody wanted to use it.
That may change soon if the company can dominate large language model-based searches. Bing’s AI is perhaps even more impressive than ChatGTP. It’s up-to-date (unlike OpenAI’s tool which is only trained on data to 2021), and provides citations for all the facts it includes in its answers, letting users double-check for accuracy.
Of course, it’s still a machine and, therefore, not perfect. But it is so good now that millions are already signing up to use it in beta. Bing never saw so much attention directed toward its services.
Is Google Really Under Threat?
OpenAI probably doesn’t pose a direct threat to Google. That’s because it has a different mission: to create artificial general intelligence (AGI). It isn’t looking to monetize web searches. Instead, it wants to build a mind comparable or superior to human intelligence.
As such, OpenAI probably won’t go up against Google. ChatGTP is clearly a helpful tool using large language models, but its search ambitions seem limited. OpenAI doesn’t appear to be angling toward that market at all. The sheer power of the tool just makes it look that way.
The real competition is from Microsoft. It has similar goals to Google and would love to be more involved in the search market. True Google-style dominance could add another trillion dollars to its market cap, benefiting shareholders tremendously.
Microsoft already tried to disrupt the search market with its 2016 chatbot, Tay. The software was pretty good even back then, but unfortunately fell foul of the mob and started spewing out all sorts of racism and bigotry.
Now, though, a ChatGTP-powered Bing using a slightly different technology promises to correct this error. Microsoft is now in a position to provide a more compelling search experience than Google with a system that can answer virtually any question and chat with users in real time.
Microsoft also faces lower reputational risk in the search market than Google. Bing only has a less than 5 percent market share, while Google’s is over 90 percent. Even if Microsoft messes up, it probably won’t affect its overall brand too much. Therefore, it can tolerate more danger than its rival.
Given the present tone of the conversation coming out of Google relating to Bard, it seems highly likely that the search giant will see significant disruption of its business model over the coming two years. Jeff Dean, for instance, has become quoted as saying that Google wants to “get it right.” Unfortunately, the disruptive tools are already here, meaning the search giant simply can’t afford to wait. It needs to take the plunge and radically change its services or risk being left behind.
How Google Can Recover
Right now, Google has limited options. However, there are a couple of ways it might contain the status quo.
The first is to become a champion of fighting misinformation online. Instead of developing user-facing AIs, it could put them to work in the backend, ensuring the public only receives truthful information.
This approach is risky because of the sheer utility of Microsoft’s tool. However, Google could potentially market itself as the “search engine of record,” similar to the strategy the New York Times uses in the media space.
It could also become the new Wikipedia, a website now mired in controversy for incorrect, inaccurate, and sometimes downright false articles. If Google can enshrine trust and become a source of truth, it will succeed.
Another approach is to maintain traditional search engine optimization by outlawing AI-generated text on the platform. If Google can develop an AI to root out machine-written text, then it can potentially maintain the current approach to ranking websites and organic traffic.
Google’s main hope is that large language models are ill-equipped to conduct genuine fact-finding. Developers can install ad-hoc guardrails, but it will always require improvisation. Users may not trust the answers they get, even if Bing backs them with various references. Machines cannot determine if something is factual or truthful, only that it is statistically likely, given their inputs.
What Will Happen If Large Language Models Take Over Search?
The effects of large language models overtaking search will be tremendous. Organic traffic to websites could potentially evaporate, annihilating SEO. In its place, we could have a machine database taking inputs from existing websites and other sources to provide users with the answers they require.
Such technology, however, risks jeopardizing the entire ecosystem. Google, Microsoft, and other search engine providers risk destroying their “seed crop,” the pages they pillage to extract data to feed their AIs. Without the incentive for new content, AI-driven tools could become progressively less capable over time.
With that said, it’s not clear how the technology might develop. If machines develop a genuine capacity for learning and knowledge storage, then they may be able to make inferences based on existing data or new information manually fed by the tech companies themselves.
Technology
Now That Large Language Models Are Here, Is Google Screwed?
